The Effects of Monetary and Exchange Rate Policy Shocks ... · PDF fileUribe, Juan Pablo...

The Effects of Monetary and Exchange RatePolicy Shocks: Evidence from an Emerging

Market Economy∗

Yasin Kursat Onder and Mauricio Villamizar-Villegasa

aCentral Bank of Colombia

Many central banks that have opted for monetary auton-omy have also been reluctant to relinquish control over thevalue of their currencies. As a result, they have operatedthrough both interest rate and foreign exchange interventions.Using daily data from the Central Bank of Turkey duringthe period of 2002–10, we study the effects of simultane-ous policies by first purging the intended monetary decisionsfrom responses to real-time macroeconomic variables and thendetermining their impact on economic activity. We find thatthe Central Bank of Turkey adjusted its policy rate mostly inresponse to inflation levels relative to both the yearly targetand agents’ expectations, and conducted purchases and sales offoreign currency in response to exchange rate behavior. Theseresponses varied depending on whether interventions were pre-announced. In terms of effectiveness, we find that unannouncedpurchases of foreign currency had a significant effect in reduc-ing exchange rate volatility but appeared to have no effect onexchange rate changes. Announced interventions, on the otherhand, did have a significant impact on exchange rate changesand volatility. Finally, we find that changes in the policy rateaffected inflation and output growth, with a lag delay of fourand two quarters, respectively.

JEL Codes: E43, E52, E58, F31.

∗We are grateful to an anonymous referee for greatly improving the technicalaspects of our estimation strategy. We also thank Hernando Vargas, Jose DarıoUribe, Juan Pablo Zarate, Guido Kuersteiner, Behzad Diba, and Mark Huggetfor useful comments. The views expressed herein are those of the authors andnot necessarily those of the Banco de la Republica (Central Bank of Colombia).Author e-mails: [email protected]; [email protected].

159

160 International Journal of Central Banking January 2018

1. Introduction

The corner (or bipolar) hypothesis and the fix-or-float proposi-tion postulate that countries tend to move away from intermedi-ate exchange rate regimes towards either hard pegs or fully flexiblerates.1 These concepts, which became conventional wisdom through-out the beginning of the 1990s, began to lose popularity after theEast Asia crises of 1997–98 and the failure of Argentina’s currencyboard in 2001. Since then, central banks have allegedly opted formonetary policy autonomy but have been reluctant to relinquishcontrol over the value of their currencies. In fact, countries under aninflation-targeting regime have led concerted initiatives to affect thevalue of major currencies, some of which include the SmithsonianAgreement (1971), the Plaza and Louvre Accords (1985, 1987), theChiang Mai Initiative (2000), and the Pittsburg Agreement (2009).

The Turkish case is no exception. Following the 2001 crisis,the Turkish economy underwent a structural transformation. TheCentral Bank of the Republic of Turkey (CBRT henceforth) wasvested with independence and endowed with the primary objectiveof achieving and maintaining price stability. In 2002, the CBRT offi-cially adopted an inflation-targeting regime and managed to bringhigh and chronic inflation down to single digits. Concurrently, inorder to lower exchange rate volatility, the CBRT conducted foreignexchange interventions in one of two ways: (i) through unannouncedinterventions, often infrequent but large, and (ii) through announcedinterventions which consisted of predetermined dates and amounts,although with a discretionary (optional) amount of trading thattook place during the day of the auction provided that monetaryauthorities decided to exceed the established amount.

In this paper we study the impact of simultaneous central bankpolicies in a unified framework. Namely, we analyze the effectsof both interest rate and foreign exchange intervention on severalmacroeconomic variables that include inflation, output growth, andexchange rate behavior. Hence, our main objective is to evaluate theeffectiveness of various types of central bank intervention. To thisend, we extend the framework presented in Romer and Romer (2004)to allow for a bivariate policy model in which policy decisions are

1See Eichengreen (1994) and Obstfeld and Rogoff (1995).

Vol. 14 No. 1 The Effects of Monetary and Exchange Rate 161

governed by dependent decision processes. Specifically, we model theundertakings of monetary authorities (tailored to the various foreignexchange mechanisms of the CBRT) using a parametric approach,and purge the intended monetary decisions from responses to high-frequency and real-time macroeconomic data. Hence, an advantageof our estimation is that it allows for non-linearities when extractingthe unexpected component of policy.

A key feature of our identification strategy consists of matchingthe actions of monetary authorities with stated targets and observ-able covariates—in other words, to closely observe what monetaryauthorities observed, and to capture their direct undertakings, espe-cially with a clear timing profile. To this end, we employ proprietarydata from the CBRT, comprising all direct sales and purchases offoreign currency as well as changes in the policy rate. We note thatour measure of the policy rate differs from any market-based rate(such as the interbank rate) in the sense that it more accuratelycaptures the intended decisions of the CBRT.2

To date, empirical studies have yet to converge on the effectsof foreign exchange intervention. For instance, studies by Menkhoff(2013) or Villamizar-Villegas and Perez-Reyna (2017) show thatnearly half of the surveyed literature find non-significant or incon-clusive results. And studies that do find a significant impact mostlyconclude that exchange rate effects are small and short-lived (see,for example, Fatum and Hutchison 1999 and Neil and Fillion 1999).

Furthermore, studies that center on the Turkish economy arerather limited, and some even face the challenge of coveringrestricted periods in which interest rate cuts always preceded pur-chases of foreign currency, making it harder to disentangle policy-specific effects. Akinci et al. (2006), for example, study eleven directintervention episodes during 2001–03 using a time-varying parame-ter model to analyze the effects on curbing exchange rate volatil-ity. Guimaraes and Karacadag (2004) also study the effects onexchange rate levels and volatility during the same time frame, usinga GARCH model. In turn, Kilinc and Tunc (2014) use a structural

2In several countries, including the United States, a researcher has to some-times infer the intended policy rate with the use of narrative records (see Romerand Romer 2004). In other cases, studies simply use overnight market rates (seeKilinc and Tunc 2014).


VAR to study the effects of policy on the Turkish economy during2006–13.

The study that most closely relates to ours is Herrera and Ozbay(2005), which studies central bank interventions using a dynamiccensored regression model during 1993–2003. In contrast, our papermainly focuses on the effects of simultaneous policies. Hence, theunified policy framework provided in this study makes our workmore amenable for empirical analyses and enables us to control forvarious policy interactions. Additionally, we differ in our definitionof censored interventions and we only focus on the time period inwhich the CBRT adopted an inflation-targeting regime.3

Our investigation confirms some of the previous findings from theliterature, but also yields some new results. Similar to Villamizar-Villegas (2016), we find that the price puzzle (i.e., positive relation-ship between prices and the policy rate) disappears once monetaryshocks are purged from systematic responses of policy. Also, in linewith Romer and Romer (2004), we find that a monetary contractionlowers industrial output with a one-quarter lag delay. On the otherhand, we find that unannounced purchases of foreign currency had asignificant effect in reducing exchange rate volatility, but appearedto have no effect on exchange rate changes. This result is similar tothose found in Dominguez (1993), Bonser-Neal and Tanner (1996),Baillie and Osterberg (1997), Chang and Taylor (1998), Fatum andHutchison (2003), Domac and Mendoza (2004), and Humala andRodriguez (2010). However, in contrast with this strand of the liter-ature, we find that announced interventions can affect both exchangerate changes and volatility.4

To the best of our knowledge, only a handful of studies existthat directly address the issue of having multiple policy instruments,few of which estimate their dependence. In this sense, we believethat our investigation will shed some light on pressing monetarypolicy questions such as the following: Under what conditions do

3While Herrera and Ozbay (2005) treat all episodes of no intervention ascensored, we consider a variety of different censoring scenarios, all of which arepresented in section 2.4. As a result, we find that announced and unannouncedinterventions have different policy implications.

4A few authors, such as Neil and Fillion (1999), Kearns and Rigobon (2002),Fatum and Hutchison (2003), and Rincon and Toro (2010), also find a significant(albeit short-lived) effect on the exchange rate.


central banks intervene in the foreign exchange market? What arethe effects of having multiple instruments? How long do these effectslast? And, finally, are decisions about various policies conducted inan independent manner?

This paper proceeds as follows: Section 2 describes the dataand emphasizes the various policy instruments undertaken by theCBRT. It also comments on the potential types of interventions thatcould have been censored by external factors. Section 3 presents themethodology, tailored to the different foreign exchange interventionmechanisms. Section 4 presents the results and section 5 concludes.

2. Data and Context

Our data cover the period of January 2002 through May 2010. Thistime frame was particularly chosen since, prior to 2002, a fixedexchange rate regime was established. Following the 2001 crisis, theTurkish economy underwent a structural transformation. The CBRTwas vested with independence and endowed with the primary objec-tive of achieving and maintaining price stability. In 2002, the CBRTofficially adopted an inflation-targeting regime and managed to bringhigh and chronic inflation down to single digits.

Following the quantitative easing (QE) program, and in orderto address challenges posed by excess capital volatility, the CBRTadopted a set of additional monetary instruments in the second halfof 2010, including a reserve option mechanism (ROM) and an inter-est rate corridor.5 Consequently, we feel that further assumptionsare needed after May 2010 in order to disentangle the effects of thenewly established tools on both the interest rate and the Turkishlira.

2.1 Foreign Exchange Interventions

In order to lower exchange rate volatility, the CBRT conducted for-eign exchange interventions in one of two ways: (i) through unan-nounced interventions, exercised through direct sales and purchasesof USD, and (ii) through announced interventions, consisting of

5See Kara (2013) for a review of unconventional monetary measures under-taken by the CBRT.


Figure 1. Different Types of Foreign ExchangeInterventions Conducted by the CBRT

Notes: The left panel corresponds to sales and purchases of USD (in millions)through announced auctions along with optional purchases. The right panel cor-responds to direct interventions. During the time of the study, it is clear thattotal purchases exceeded total sales.

predetermined dates and amounts, although with a discretionary(optional) amount of trading that took place during the day of theauction provided that monetary authorities decided to exceed theestablished amount.6 Announcements for this last type of interven-tions ranged from one day to two weeks prior to the currency auction.Also, the CBRT did not use a deterministic rule to decide over whichdate to intervene or the amount of currency to be traded.

The left panel of figure 1 depicts the total number of sales andpurchases (in millions of USD) through announced auctions alongwith optional purchases of foreign currency (the CBRT never con-ducted optional sales). As shown, the CBRT purchased foreign cur-rency throughout most of the sample, with optional purchases start-ing in September 2003, and occasional sales during 2006, 2008, and2009. Alternatively, the right panel of figure 1 depicts the totalnumber of sales and purchases (in millions of USD) through unan-nounced interventions. Under this type of trading, purchases andsales were infrequent but large, averaging $1.7 and $0.3 billion USD,respectively.

6There were limits on how much the CBRT could exceed the pre-establishedamount.


Similarly, table 1 shows the amount of foreign currency traded forevery type of foreign exchange intervention. As seen, all purchaseswere larger than sales, by more than tenfold. As will be discussed ingreater depth in the next section, this asymmetry reveals a system-atic bias towards trying to depreciate domestic currency. As such,we address the fear of floating or rather, the fear of appreciation, byallowing some of these interventions to follow a censored Tobit typeI model. Table 1 also shows that sales were largely concentrated inthe year 2006. Purchases, on the other hand, were most abundantin 2003 and 2005.

2.2 Policy Rate

Our measure of the policy rate corresponds to the CBRT’s overnightborrowing rate between February 20, 2002 and May 16, 2008 (due tothe abundant liquidity in the Turkish market); to the overnight lend-ing rate between May 17, 2008 and May 20, 2010 (due to the liquid-ity shortage); and to the one-week repo lending rate after May 21,2010. As such, our investigation differs from studies that use market-based rates such as the interbank rate. We argue that the lattercan be more likely influenced by monetary factors driven by liquid-ity demand, as they comprise equilibrium conditions which reflecttransactions within the financial system, including those betweencommercial banks and other non-banking entities. However, we con-duct robustness exercises (reported in figure 10 in appendix 3) inorder to assess the differences obtained if we had instead used theinterbank rate.

Figure 2 depicts our measure of the policy rate and the interbankrate (left panel), as well as the observed, targeted, and expectedyearly inflation (right panel). The figure shows that at the onset of2002, inflation (or hyperinflation) levels reached 73.2 percent, whilethe target for inflation was set at 35 percent and the policy rate(depicted in the left panel) was set at 59 percent. With a sharpdisinflation in 2001, the policy rate began to steadily decline untilmid-2006. Starting in mid-2006, interest rates slightly reboundedbut started falling again in 2008.

The positive relationship between inflation and interest rates canbe misconstrued as evidence of the price puzzle. Nonetheless, a morereasonable explanation was that the CBRT kept interest rates high


Tab

le1.

For

eign

Exch

ange

Inte

rven

tion

sJa

nuar

y20

02–M

ay20

10(b

illion

USD

purc

has

es)

USD

Tot

al02

0304

0506

0708

0910

Una

nnou

nced

Pur

chas

es25

.53

0.02

4.23

1.28

14.5

75.

440.

000.

000.

000.

00U

nann

ounc

edSa

les

2.12

0.01

0.00

0.01

0.00

2.11

0.00

0.00

0.00

0.00

Ann

ounc

edP

urch

ases

28.5

20.

794.

992.

603.

632.

244.

753.

602.

913.

00A

nnou

nced

Sale

s2.

000.

000.

000.

000.

001.

000.

000.

10.

900.

00O

ptio

nalP

urch

ases

20.4

10.

000.

661.

503.

812.

115.

153.

981.

401.

78

Sourc

es:C

entr

alba

nkda

taan

dau

thor

’sca

lcul

atio

ns.


Figure 2. Policy Rate and Yearly Inflation(observed, target, and expected)

Notes: The left panel corresponds to the CBRT policy rate. The right panelcorresponds to the observed inflation, to the surveyed one-year-ahead expectedinflation forecast, and to the yearly target.

to bring inflation down to single digits, and only conducted expan-sionary monetary policy once inflation decreased. In section 4.4 weshow that the price puzzle is eliminated by purging the policy ratefrom systematic responses of policy.

2.3 Simultaneous Policies: Leaning with or against the Wind?

Figure 3 depicts episodes in which the CBRT conducted announcedand unannounced purchases and sales of foreign currency, along withchanges in the policy rate. In the left panel, the solid (dashed) linesdenote unannounced purchases (sales) of foreign currency. Hence,it shows that purchases were used in tandem with the policy rate,as they were conducted during episodes of interest rate cuts. How-ever, as shown by the blue dashed lines,7 there were some episodesin 2002 and 2004 in which the CBRT conducted leaning-against-the-wind policies (i.e., interest rate cuts along with sales of foreigncurrency).

Similarly, the right panel of figure 3 shows that in 2008 and2009, announced sales (orange lines) coincided with interest rate

7Colors appear in the online version, available at http://www.ijcb.org.


Figure 3. Simultaneous Monetary Policies

Notes: The left panel corresponds to unannounced interventions, where solid(dashed) lines denote purchases (sales) of foreign currency. Some purchases werenot included given that they did not occur within one week of a policy ratechange. The right panel corresponds to announced interventions, where shadedregions (solid lines) denote purchases (sales) of foreign currency.

cuts, exerting potentially opposing forces on the exchange rate.8

Lastly, given the high number of announced purchases throughoutthe sample, they coincided with both interest rate hikes (leaningagainst the wind) and interest rate cuts (leaning with the wind).

2.4 Censored Interventions

Earlier we highlighted the asymmetry between purchases and sales offoreign currency conducted by the CBRT. That is, while purchasestotaled 74.5 billion USD, sales totaled only 4.12 billion USD (seetable 1). When modeling the various policy functions for foreignexchange intervention, the general absence of USD sales can takeon two different interpretations: (i) either economic conditions weresuch that it was optimal for the CBRT to conduct only purchasesof USD, or (ii) economic conditions were such that it was optimalfor the CBRT to conduct sales of USD, but it did not carry themout given some external factor or constraint. The latter describes a

8Leaning-against-the-wind policies of both announced and unannounced for-eign exchange interventions generally took place during heightened global finan-cial market volatility, as can be seen in figure 9 of appendix 2.


Table 2. Foreign Exchange InterventionsConsidered as Censored

Type Censored Not Censored

Unannounced Total (Purchases – Sales) XUnannounced Purchases XUnannounced Sales XOptional Purchases XAnnounced Purchases XAnnounced Sales XChanges in Policy Rate X

censored process that, if estimated with a linear model, would yieldinconsistent estimates.9

As such, table 2 describes the different specifications consideredin this study when modeling policy. In essence, announced interven-tions and changes in the policy rate were not considered as beingcensored, while optional purchases were considered as censored giventhe complete lack of optional sales. Finally, unannounced interven-tions were considered as both censored (when taking sales and pur-chases individually) and uncensored (when taking the total). Themain reasons for allowing unannounced interventions to have bothspecifications were to establish a benchmark comparison with otherstudies that also assume censored policy processes,10 as well as toanalyze the importance of the conditional probability of observinga positive intervention, by comparing both types of estimations.

3. Methodology

3.1 Policy Effects in a Potential Outcomes Framework

The main challenge of estimating the effects of policy is that mon-etary decisions are rarely isolated from economic developments. Ina potential outcomes framework, this corresponds to not being able

9See Cohen (1949), Rosenbaum (1961), or Barr and Sherrill (1999).10See, for example, Herrera and Ozbay (2005) and Villamizar-Villegas (2016).


to properly account for the systematic differences between treat-ment and control groups (i.e., intervention versus non-interventionepisodes). For instance, assume that we are interested in the causaleffect of purchasing “s” units of foreign currency. A counterfactualof interest can be stated as the effects of purchasing “s” given thatthe central bank actually purchased “s − j.”11 However, conditionscould have significantly differed for when the CBRT purchased “s”or “s − j” (which most likely explains the difference in interventionvalues).

To see this point more formally, let Yt be a vector of outcome vari-ables, Dt a vector of policy instruments, and Xt a matrix of covari-ates needed to characterize the various policy functions. Histories ofpolicy, outcomes, and exogenous variables are characterized as

⎡⎣Dt

Xt

Yt

⎤⎦ =

⎡⎣Dt Dt−1 · · · Dt−k

Xt Xt−1 · · · Xt−k

Yt Yt−1 · · · Yt−k

⎤⎦ , (1)

and the “relevant” statistic that policymakers use to determine pol-icy at time “t” can be described by zt = Φt(Yt, Xt, Dt−1), for a givenmapping Φt.

In principle, differences in outcome variables, whenever theCBRT purchased “s” units of foreign currency compared with whenit purchased “s − j,” can be formulated as follows:

E [Yt,s | Dt = s] − E [Yt,s−j | Dt = s − j] = (2)

E[Yt,s − Yt,s−j | Dt = s] + E [Yt,s−j | Dt = s]

− E [Yt,s−j | Dt = s − j] , (3)

where Yt,s corresponds to the vector of potential outcomes had thebank purchased “s” units of foreign currency, regardless of the actualamount purchased. Alternatively, observed purchases are dictatedby the realization of Dt. The step between equations (2) and (3)simply corresponds to the addition and subtraction of the termE [Yt,s−j | Dt = s], that is, the conditional mean of Yt had the CBRTpurchased “s − j” units of foreign currency when it in fact pur-chased “s.”

11Note that “j” can be set equal to “s” for the case of no interventions.


Equation (3) is hence comprised of three terms. The first termis our variable of interest and captures the average treatment effect(ATE) of purchasing “j” additional units of foreign currency (from“s − j” to “s”), given a purchase of “s.” The second and thirdterms constitute the resulting bias that arises due to the non-randomization of treatment assignment.12

Fortunately, the conditional independence assumption (CIAhenceforth) allows us to eliminate this bias. Namely, the CIA statesthat conditional on the relevant history, policy decisions are inde-pendent of potential outcomes, or as good as randomly assigned.This assumption, sometimes known as the selection-on-observablesassumption, establishes the foundation based on which “regressionscan also be used to approximate experiments in the absence of ran-dom assignment.”13 Formally, the CIA can be stated as equation(4):

Yt,s(dt) ⊥ Dt | zt ∀d ∈ D,∀s. (4)

Consequently, the bias that corresponds to the second and thirdterms of equation (3) cancels out as shown:

BIAS = E [Yt,s−j | zt, Dt = s] − E [Yt,s−j | zt, Dt = s − j]

= E [Yt,s−j | zt, Dt = s] − E [Yt,s−j | zt, Dt = s]

= 0. (5)

Equation (5) follows from the fact that policy instruments areindependent of potential outcomes. As such, it is essential to extractthe random component of policy from anything that may systemat-ically react to informative variables. In the empirical application,it justifies the two-step procedure of first identifying exogenousmonetary shocks and then estimating their effects on the economy.Accordingly, the first step of our methodology consisted of modelingthe various policy rules in order to remove systematic responses toinformative variables.

12For example, when evaluating the effects of policy on inflation, it is rea-sonable to argue that the bias in equation (3) will most likely be positive,E [Yt,s−j | Dt = s] > E [Yt,s−j | Dt = s − j], falsely attributing a greater effecton policy.

13Angrist and Pischke (2009, p. 18).


3.2 Computation of Monetary Shocks

3.2.1 Uncensored Policies

When computing monetary shocks, there is no reason to believe thatpolicy decisions were independent. After all, the CBRT conductedmonetary policy through both foreign exchange interventions (FXIhenceforth) and interest rate interventions (IRI henceforth), and itis entirely plausible that decisions about one instrument altered theprobability distribution of the other.14 We thus proceed by parame-terizing this dependence as follows:

FXI∗t = x

′

1tβ1 + ε1t

IRIt = x′

2tβ2 + ε2t(ε1t

ε2t

)∼ N (0, Σ) , (6)

where the residuals of both policy functions (i.e., policy shocks)are assumed to be jointly normal with zero mean and variance-

covariance matrix Σ =[σ2

1 σ12σ12 σ2

2

].

The construction of a maximum-likelihood function for thebivariate process described in equation (6) is hence warranted inorder to obtain estimates of all individual regressors as well asthe estimated covariance between policy shocks. The correspondingbivariate normal likelihood is presented as

Ln(θ) =T∏

t=1

1

(2π) |Σ|1/2 e− 12 (Dt−μ)′Σ−1(Dt−μ), (7)

where Dt =[FXIt

IRIt

]and μ = Et

[x

′

1tβ1

x′

2tβ2

]. If the estimation of the

maximum likelihood yields a significant covariance between policyresiduals (σ12), shocks can then be computed in vector form in orderto account for the conditional dependence of policy. Otherwise, theycan be computed by estimating linear fitted residuals of independentprocesses characterized by equation (6).

14In fact, the left panel of figure 3 suggests that both interest rate and foreignexchange interventions were sometimes orchestrated.


3.2.2 Censored Policies

Table 2 denotes which types of FXI were considered as censored andthus modeled with a Tobit type I model. We proceed to parameter-ize the maximum-likelihood function for the bivariate policy processin which FXI exhibit some degree of censoring.

Formally, let A ≡(σ2

1 − σ212

σ22

)and b ≡

(x

′

1tβ1+σ12σ22

(IRIt−x′

2tβ2)).

It follows that

Ln(θ) =∏

FXI∗t ≤0

f (FXIt, IRIt | x1t, x2t)

×∏

FXI∗t >0

f (FXIt, IRIt | x1t, x2t)

=∏

FXI∗t ≤0

(1 − Φ

(b

A1/2

))1σ2

φ

(IRIt − x

′

2tβ2

σ2

)

×∏

FXI∗t >0

1A1/2 φ

(FXI∗

t − b

A1/2

)1σ2

φ

(IRIt − x

′

2tβ2

σ2

), (8)

where φ(·) and Φ(·) correspond to the probability distribution func-tion and the cumulative distribution function of a standard normaldistribution, respectively. Similar to the case of uncensored poli-cies, if the estimation of the maximum likelihood yields a significantcovariance between policy residuals (σ12), they can be computed invector form as presented in Villamizar-Villegas (2016). However, ifthe covariance is not significant, policy shocks can be obtained bysubtracting the conditional mean of policy from its observed value,as follows:

ε1t = FXIt − E [FXIt | x1t]

= FXIt − Φ

(x

′

1tβ1

σ1

) [x

′

1tβ1 + σ1λ

(x

′

1tβ1

σ1

)](9)

ε2t = IRIt − E [IRIt | x2t]

= IRIt − x′

2tβ2, (10)


where the term λ(·) = φ(·)/Φ(·) corresponds to the inverse Mills

ratio. The term Φ(

x′1tβ1σ1

)of equation (9) represents the probabil-

ity of observing a positive intervention (i.e., Pr(FXI∗t > 0 | x1t)),

whereas the last term in brackets is the expected value of the latentvariable FXI∗

t .15

3.3 Impulse Response Functions

The second step of the methodology consisted of estimating theeffects of the estimated residuals, ε1t and ε2t, on the different out-come variables in Yt. To this end, we estimated impulse responsefunctions (IRFs) for variables with a monthly frequency accordingto Jorda’s (2005) methodology of local projections:

Yit+s = ηs0 + ηs

1ε1t + ηs2ε2t + ϑit+s for s = 0, 1, . . . , h. (11)

In this case, the correlation between policy lags disappears sinceshocks are summed up into monthly observations. Conversely, weestimated IRFs for variables with a daily frequency according to themethodology of Romer and Romer (2004):

Yit = γ0 +h∑

j=0

γjε1t−j +h∑

k=0

γkε2t−k + ςit. (12)

Coefficients and standard errors (bootstrapped) were summed upevery period in order to obtain the cumulative effect across time.16

4. Estimation and Results

4.1 Parametric Dependence of Monetary Shocks

Estimation results for the maximum-likelihood function of equa-tions (7) and (8) are reported in table 3. Values correspond to the

15Residuals ε1t and ε2t correspond to the policy shocks for FXI t and IRI t,respectively.

16The number of lags varied depending on the frequency of the outcome vari-able (h = 12 if monthly, h = 40 if daily). IRFs were smoothened using a movingaverage of ± two lags, for readability purposes only.


Table 3. Covariances of Bivariate Process

Bivariate Model Uncensored Censored

FX Unannounced Total – Policy Rate −0.05(0.034)

FX Announced Purchases – Policy Rate −0.02 −0.06(0.034) (0.041)

FX Announced Sales – Policy Rate 0.08∗∗ 0.65∗∗∗

(0.034) (0.208)FX Unannounced Purchases – Policy Rate 0.40

(0.289)FX Unannounced Sales – Policy Rate 0.10

(0.205)FX Optional Puchases – Policy Rate −0.08

(0.064)

Notes: All models consisted of 2,190 observations. Log-likelihoods for each model cor-respond to 1492.2, 8498.65, and 8592.0 for the uncensored specification, respectively,and to 875.1, 889.4, 5234.1, and 3185.1 for the censored specification, respectively.*, **, and *** indicate significance at the 10 percent, 5 percent, and 1 percent level,respectively. Standard errors are reported in parentheses.

covariance between vt and ε2t. As can be observed, none of the covari-ances are statistically significant except for the covariance betweenannounced sales and the policy rate. For computational purposes,policy shocks were estimated according to equation (9) (for cen-sored observations of announced purchases, unannounced purchasesand sales, and optional purchases); equation (10) (for uncensoredobservations of policy rate changes); fitted residuals equivalent to

when Φ(

x′1tβ1σ1

)= 1 in equation (10) (for uncensored observations

of unannounced totals, and announced purchases and sales); andequation (7) (for announced sales) when computing shocks in vectorform.

Interestingly, most of these covariances turn significant whenexcluding inflation values relative to the yearly target from Xt (allexcept unannounced trades whose covariance is not significant acrossthe different specifications of Xt). These findings indicate that, underthe assumptions of the model, the CBRT’s decisions of one instru-ment (conditional on Xt) did not alter the probability distribution


of the other. However, this result does not mean that different poli-cies did not react to the same target. In fact, many covariates thatwere included in x1t were also included in x2t.17 Independence, inthis case, is conditional on the set of control variables.

4.2 Policy Functions

Small open economies, such as Turkey, are known to be vulnera-ble to global financial conditions. As such, it is important to con-trol for variables that affect both the financial and macroeconomiccycle. As Rey (2015, p. 21) states, “Fluctuating exchange rates can-not insulate economies from the global financial cycle, when capitalis mobile. The ‘trilemma’ morphs into a ‘dilemma’—independentmonetary policies are possible if and only if the capital accountis managed, directly or indirectly, regardless of the exchange-rateregime.” We thus proceed by including variables such as the VIXindex and the SPG Commodity Channel Index (SPGCCI) in theestimations that follow. All other variables included in our estima-tions are described in appendix 1, and their stationarity propertiesare reported in table 10 of appendix 4.18

Tables 4–7 show results for all the various types of foreignexchange policy functions. Table 4 shows that unannounced totalinterventions mostly reacted to the exchange rate behavior (meas-ured in Turkish lira per U.S. dollar (TRY/USD)). That is, theCBRT tried to depreciate domestic currency by purchasing USDwhenever the exchange rate appreciated (relative to the daily,weekly, and monthly exchange rate) and whenever exchange ratevolatility increased. In specifications 3 and 4, interventions posi-tively responded to lagged announced purchases of USD (exhibit-ing some momentum effect) as well as to changes in the policyrate (specification 3), indicating a leaning-against-the-wind policy.19

17An example is lagged interest rate interventions (ΔIRIt−1), which wereincluded in all specifications.

18We are grateful to an anonymous referee for suggesting these variables asproxies of global financial conditions. We also conducted additional exercises (notreported) using changes in the oil price (BRENT) instead of the SPGCCI index,yielding similar results.

19The CBRT purchased foreign currency while simultaneously conducting con-tractionary monetary policy.


Table 4. OLS Estimation: FXI totalt = x ′

1tβ1 + υt

FX Unannounced Total(Uncensored) (1) (2) (3) (4)

FXI totalt−1 0.00 0.00 0.00 0.00

(0.010) (0.010) (0.009) (0.009)Inflation(πt−1) – Target(π∗

t−1) −8.01 −8.53 −11.10 −11.09(9.582) (9.670) (10.131) (10.064)

Industrial Production ΔIndt−1 0.68∗ 0.73∗ 0.54 0.31(0.368) (0.376) (0.383) (0.421)

Daily Depreciation Δet−1 −4.09∗ −4.66∗∗ −4.68∗∗∗ −6.07∗∗

(2.102) (2.202) (2.195) (2.824)Weekly Depreciation Δet−5 −2.07∗ −2.37∗∗ −2.70∗∗ −2.95∗∗

(1.126) (1.192) (1.225) (1.264)Monthly Depreciation Δet−20 −0.85∗ −1.08∗∗∗ −0.90∗∗ −0.55

(0.491) (0.668) (0.466) (0.546)Exchange Rate Vol t−1 1.41∗∗ 1.60∗∗ 1.69∗∗

(0.669) (0.694) (0.669)Announced Purchasest−1 0.51∗∗∗ 0.51∗∗∗

(0.185) (0.186)Inflation Surprises (πt−1 − πe

t−1) 0.00 0.05(0.326) (0.300)

Policy Rate ΔIRIt−1 3.70∗ 3.18(2.207) (2.108)

VIXt−1 −0.42(0.396)

Commodity Index ΔSPGCCIt−1 −0.87∗

(0.523)

Notes: Specifications x1t(1–4) correspond to the different combinations of covari-ates. FX intervention is measured in millions of USD. All specifications consistedof 2,190 observations. R2 = 0.005, 0.005, 0.007, and 0.007 for OLS specifications1–4. *, **, and *** indicate significance at the 10 percent, 5 percent, and 1 percentlevel, respectively. Standard errors are reported in parentheses. Constant and yeardummies not reported.

Additionally, specifications 1 and 2 show a positive effect of indus-trial output growth on USD purchases.20 Finally, specification 4shows that the CBRT purchased USD whenever commodity prices

20The effect of industrial output is not significant in specifications 3 and 4, pos-sibly due to the high correlation between lagged output growth and the policyrate.


Table 5. Tobit Estimation: FXI t = max [0,x ′1tβ1 + υt ]

Unannounced FXI (Censored) Purchases Sales

Inflation(πt−1) – Target(π∗t−1) −1.87 0.27

(1.233) (0.359)Industrial Production ΔIndt−1 0.13 0.04

(0.095) (0.036)Weekly Depreciation Δet−5 −3.23∗ 0.69∗

(1.654) (0.417)Monthly Depreciation Δet−20 −0.80∗∗∗ 0.08∗

(0.303) (0.040)Exchange Rate Vol t−1 −0.39 −0.22

(0.761) (0.143)Policy Rate ΔIRIt−1 0.50 −0.08

(1.614) (0.499)VIXt−1 −0.36∗∗ 0.00

(0.174) (0.018)Commodity Index ΔSPGCCIt−1 −0.10 0.00

(0.099) (0.032)

Notes: Specifications x1t(1–2) correspond to the different combinations of covari-ates. FX intervention is measured in millions of USD. All specifications consisted of2,190 observations. Pseudo R2 = 0.12 and 0.23 for Tobit specifications 1–2. *, **,and *** indicate significance at the 10 percent, 5 percent, and 1 percent level, respec-tively. Standard errors are reported in parentheses. Constant and year dummies notreported.

(SPGCCI) decreased. Similarly, table 5 shows results for censoredunannounced sales and purchases, when considered individually.Results are similar to those of table 4: purchases (sales) followedappreciating (depreciating) exchange rate episodes. One difference,nonetheless, is the significant effect of the VIX index on unan-nounced purchases.21

Table 6 presents the estimation results for optional purchases(recall that this type of FXI was considered as censored, given thecomplete lack of optional sales). Surprisingly, results show that many

21Tables 4–7 show that responses of the CBRT to global financial conditions(proxied by the VIX index) were addressed through either unannounced andoptional purchases or through announced sales.


Table 6. Tobit Estimation: FXIoptionalt = max [0,x ′

1tβ1 + υt ]

FX Optional Purchases(Censored) (1) (2) (3) (4)

FXI optionalt−1 0.93∗∗∗ 0.93∗∗∗ 0.65∗∗∗ 0.62∗∗∗

(0.048) (0.048) (0.048) (0.048)Inflation(πt−1) – Target(π∗

t−1) −0.52∗∗∗ −0.51∗∗∗ −0.43∗∗∗ 0.67∗∗∗

(0.143) (0.143) (0.144) (0.164)Industrial Production ΔIndt−1 0.53∗∗∗ 0.51∗∗∗ 0.28∗∗∗ 0.03

(0.082) (0.083) (0.083) (0.097)Monthly Depreciation Δet−20 −0.60∗∗∗ −0.48∗∗ −0.44∗∗ −0.02

(0.190) (0.203) (0.202) (0.213)Exchange Rate Vol t−1 −0.64 −0.53 −0.36

(0.399) (0.391) (0.399)Announced Purchasest−1 0.70∗∗∗ 0.72∗∗∗

(0.054) (0.055)Unannounced Purchasest−1 0.00 0.00

(0.003) (0.003)Policy Rate ΔIRIt−1 3.76 2.92

(3.000) (2.918)VIXt−1 −0.72∗∗∗

(0.111)Commodity Index ΔSPGCCIt−1 0.03

(0.095)

Notes: Specifications x1t(1–4) correspond to the different combinations of covari-ates. FX intervention is measured in millions of USD. All specifications consisted of2,190 observations. Pseudo R2 = 0.06, 0.06, 0.07, and 0.11 for Tobit specifications1–4. *, **, and *** indicate significance at the 10 percent, 5 percent, and 1 percentlevel, respectively. Standard errors are reported in parentheses. Constant and yeardummies not reported.

variables affected the decision to optionally intervene in the foreignexchange market, as if expected by market participants.22 In thiscase, the CBRT tried to depreciate domestic currency by purchas-ing USD whenever inflation was low (relative to the yearly target),whenever industrial output increased, and whenever the monthly

22This can be explained by the numerous times that the CBRT conductedoptional purchases. In fact, after September 2003, the CBRT almost alwaysexceeded the pre-established amount.


Table 7. OLS Estimation: FXI t = x ′1tβ1 + υt

Announced FXI (Censored) Purchases Sales

FXIt−1 0.82∗∗∗ 0.16(0.021) (0.139)

Inflation(πt−1) – Target(π∗t−1) 1.34∗∗∗ 0.39∗∗

(0.354) (0.167)Industrial Production ΔIndt−1 0.06∗∗∗ −0.08∗∗∗

(0.017) (0.019)Monthly Depreciation Δet−20 −0.11∗∗ 0.03

(0.048) (0.059)Exchange Rate Vol t−1 −0.05 −0.05

(0.065) (0.058)Unannounced Purchasest−1 0.002∗ 0.00

(0.001) (0.001)Unannounced Salest−1 −0.002 0.47∗∗∗

(0.002) (0.132)Inflation Surprises (πt−1 − πe

t−1) −.11∗∗∗ −0.04∗∗∗

(0.036) (0.012)Policy Rate ΔIRIt−1 −1.12∗∗ 0.06

(0.571) (0.998)VIXt−1 −0.01 0.04∗∗

(0.014) (0.016)Commodity Index ΔSPGCCIt−1 0.02 0.02

(0.016) (0.023)

Notes: Specifications x1t(1–2) correspond to the different combinations of covariates.FX intervention is measured in millions of USD. All specifications consisted of 2,190observations. R2 = 0.74 and 0.79 for OLS specifications 1–2. *, **, and *** indicatesignificance at the 10 percent, 5 percent, and 1 percent level, respectively. Standarderrors are reported in parentheses. Constant and year dummies not reported.

exchange rate appreciated.23 Specifications 3 and 4 also show thatinterventions positively responded to past purchases of announcedinterventions. Finally, specification 4 shows a significant and nega-tive impact of the VIX index.

23The CBRT could have reacted to inflation levels, as some authors argue thateven sterilized interventions can have an effect on prices via liquidity premiums(see Canzoneri and Cumby 2013).


Table 7 shows results for announced sales and purchases. Whilepurchases were conducted after positive changes in industrial outputgrowth, sales followed negative changes in output. Also, purchaseswere conducted whenever the monthly exchange rate appreciated,but sales seemed not to respond to exchange rate changes. Webelieve, however, that this is due to the scant number of sales withinour sample. Additionally, sales and purchases reacted to inflationchanges (relative to both the target and expected inflation).

In sum, these results are similar to those found in Kamil (2008)and Echavarria et al. (2013) in that central banks are inclined topurchase (sell) foreign currency whenever the exchange rate appre-ciates (depreciates). However, a novel feature in our investigationis that responses mostly varied depending on whether interventionswere pre-announced.

Similar to Romer and Romer (2004), the IRI t policy functionof equation (6) was estimated using OLS around meeting date “m”of the open market committee of the CBRT. This setting (like inRomer and Romer 2004) assumes that unemployment acts throughthe measure of GDP gap (i.e., Okun’s Law). Results are reportedin table 8. Coefficients of the lagged policy rate ΔIRIt−1 are smalland for the most part not statistically significant. Estimates alsoshow that the main explanatory variable was inflation relative tothe yearly target (i.e., the CBRT conducted contractionary policyin order to lower inflation). Other variables that prompted policyadjustments were lagged values of announced purchases and unan-nounced sales, weekly and monthly exchange rate changes, and theVIX index (i.e., the CBRT conducted monetary easing when marketturmoil increased). We note that the negative impact of inflationsurprises on policy adjustments is conditional to the inclusion ofinflation changes with respect to the target rate. In many specifica-tions, the effect of inflation surprises is reversed when excluding allother measures of inflation. The decision to include both measuresof inflation was justified by their medium-to-low correlation of 0.32.In all specifications, the CBRT did not seem to respond to changesin the U.S. federal funds rate nor industrial output.

4.3 Policy Shocks

Figure 4 depicts the resulting monetary shocks (ε1t, ε2t) comparedwith the observed policy instruments (FXI t, IRIt). To improve


Table 8. OLS Estimation: ΔIRIm = x ′2tβ2 + ε2t

Policy Rate Changes(Uncensored) (1) (2) (3) (4)

ΔIRIm−1 0.35∗ 0.24 0.20 0.12(0.202) (0.171) (0.176) (0.152)

Inflation(πt−1) – Target(π∗t−1) 0.19∗ 0.22∗∗ 0.23∗∗ 0.28∗∗∗

(0.099) (0.099) (0.103) (0.066)Weekly Depreciation Δet−5 −0.13∗ −0.19∗∗∗ −0.19∗∗∗ −0.12

(0.075) (0.064) (0.061) (0.078)Monthly Depreciation Δet−20 0.06∗ 0.06∗ 0.06∗∗ 0.07∗

(0.028) (0.033) (0.025) (0.031)Industrial Production ΔIndt−1 0.01 0.00 0.01 0.05

(0.015) (0.014) (0.017) (0.034)Inflation Surprises (πt−1 − πe

t−1) −0.15∗∗ −0.18∗∗∗ −0.19∗∗∗ −0.14(0.063) (0.058) (0.059) (0.072)

U.S. Federal Funds Rate Δi∗t−1 −1.99 −0.71 −0.51 −1.63(3.244) (3.136) (2.831) (2.865)

Unannounced Salest−1 0.003∗∗∗ 0.003∗∗∗ 0.002(0.001) (0.001) (0.001)

Announced Purchasest−1 −0.01 −0.02∗∗∗

(0.006) (0.008)VIXt−1 −0.05∗∗

(0.021)Commodity Index ΔSPGCCIt−1 0.01

(0.021)

Notes: Specifications x1t(1–4) correspond to the different combinations of covariates.All models consisted of fifty-two observations given that estimations were conductedaround meeting dates of the board of directors. R2 = 0.59, 0.59, 0.60, and 0.67 forOLS specifications 1–4. *, **, and *** indicate significance at the 10 percent, 5 per-cent, and 1 percent level, respectively. Standard errors are reported in parentheses.Constant and year dummies not reported.

readability, all foreign exchange shocks and observed values weresummed into quarterly observations. The deterministic componentof policy can be interpreted as the difference between the darkshaded and light shaded bars (green and orange bars, respectively,in the online version). As shown, policy shocks greatly differed fromobserved values, especially during certain time periods. For instance,the left panel of figure 4 shows that the CBRT would have inter-vened less in the foreign exchange market had it not been for past


Figure 4. Observed Intervention versus New Measures ofPolicy Shocks: 2002–10

exchange rate movements. Specifically, during 2002–05 and 2007–10,most foreign exchange interventions were explained by the determin-istic component of policy. In contrast, interventions in 2005 werehighly unpredictable.

In turn, the right panel of figure 4 shows that most policy changeswere explained by inflation levels. In this case, changes in inflationexplained the variation in the policy rate by almost 65 percent, onaverage.

One important characteristic of correctly specified policy shocksis their unpredictability. In other words, information prior to thepolicy change should be uncorrelated with the estimated residuals.A heuristic exercise to test for this orthogonality condition is pre-sented in table 9. Each column denotes a different estimated policyshock, whereas each row contains the different lagged policy inter-vention. Hence, policy shocks are individually regressed against allof the different types of intervention variables. Values with an “X”correspond to the variable (row) that was included under that speci-fication (column), so the policy shock is, by construction, orthogonalto that variable. As it turns out, all residuals are correctly specifiedacross the various intervention variables.

4.4 Impact on Outcome Variables

We considered four outcome variables in order to evaluate the effectsof policy: (i) exchange rate changes, (ii) exchange rate volatility, (iii)


Tab

le9.

Pol

icy

Shock

s’O

rthog

onal

ity

Con

ditio

n(ε

it=

x′ itβ

+η

it)

Policy

Shock

Unan

nounce

dU

nan

nounce

dU

nan

nounce

dO

ptional

Announce

dA

nnounce

dPolicy

Rat

ever

sus

Tota

lsP

urc

has

esSal

esP

urc

has

esP

urc

has

esSal

esC

han

ges

Lag

ged

Polici

esε1t

ε1t

ε1t

ε1t

ε1t

ε1t

ε2t

Una

nnou

nced

XX

X0.

000.

000.

000.

00Tot

als t

−1

(0.0

02)

(0.0

01)

(0.0

01)

(0.0

01)

Una

nnou

nced

XX

0.00

XX

X0.

00P

urch

ases

t−1

(0.5

32)

(0.0

01)

Una

nnou

nced

X0.

00X

0.00

XX

XSa

les t

−1

(0.0

05)

(0.0

01)

Opt

iona

l0.

240.

29−

0.02

X0.

010.

000.

06P

urch

ases

t−1

(0.2

69)

(0.2

71)

(0.2

22)

(0.0

11)

(0.0

07)

(0.1

78)

Ann

ounc

edX

−0.

23−

0.02

XX

0.00

XP

urch

ases

t−1

(0.2

67)

(0.0

31)

(0.0

09)

Ann

ounc

ed−

0.01

−0.

01−

0.03

0.00

0.00

X−

0.00

Sale

s t−

1(0

.020

)(0

.017

)(0

.025

)(0

.001

)(0

.002

)(0

.013

)Pol

icy

Rat

eX

XX

XX

XX

Cha

nges

t−1

Note

s:A

llm

odel

sco

nsi

sted

of

2,1

90

obse

rvati

ons.

*,

**,

and

***

indic

ate

signifi

cance

at

the

10

per

cent,

5per

cent,

and

1per

cent

level

,re

spec

tivel

y.Sta

ndard

erro

rs(r

obust

for

OLS)

are

report

edin

pare

nth

eses

.


changes in inflation, and (iv) industrial production growth. In allcases, the effects of both the estimated residuals (ε1t, ε2t) and theobserved policy instruments (FXI t, IRIt) were computed.24 Whilethe former consist of correctly specified monetary surprises, the lat-ter are most likely biased by anticipatory movements in the econ-omy. The comparison of both measures is thus useful in order toget a better sense of the direction and magnitude of the bias drivenby observed interventions (see equation (3)). Hence, the left panelsof figures 5–8 depict responses in outcome variables due to policyshocks. Conversely, the right panels of figures 5–8 depict responsesin outcome variables due to observed values of intervention.25

4.4.1 Exchange Rate Changes (Daily Frequency)

Panels C and E of figure 5 show significant effects of FX policyshocks on exchange rate changes. Namely, announced purchases of 1billion USD depreciated domestic currency by up to 5 percent dur-ing days 10–40 after the intervention took place. On the other hand,announced sales of 1 billion USD appreciated domestic currencyby up to 2.5 percent, but only during the first ten days followingthe intervention shock. These results are in contrast with most ofthe recent literature that find non-significant effects of FXI on theexchange rate (see Fischer 2001a, 2001b and Blanchard 2013). How-ever, similar to Rey (2015), we argue that exchange rate effects arepossible when breaking free from the monetary trilemma.26

Other FX policy shocks such as unannounced sales and purchasesof foreign currency appeared to have a null effect on exchange ratechanges.27 Finally, consistent with the theory on interest rate pari-ties, panel G shows that a 1 percent increase in the policy rate shock(IRI shock) appreciates domestic currency by 1 percent during thefirst month (days 5–25).

24Policy residuals were computed according to specification 4 of tables 4, 6,and 8, and according to the only specification presented in tables 5 and 7.

25IRFs not reported in figures 5–8 were not statistically significant.26See Villamizar-Villegas and Perez-Reyna (2017).27These results are in line with those found in Dominguez (1993), Dominguez

and Frankel (1993), Humpage (1999), Kim, Kortian, and Sheen (2000), and Taylor(2004), but are contrary to those found in Disyatat and Galati (2007) and Adlerand Tovar (2011).


Figure 5. Implied IRFs of Exchange Rate Changes (Δet)

Notes: IRFs A–F correspond to a response in exchange rate changes (%) to a 1billion USD purchase (or sale). IRFs G–H correspond to a response in exchangerate changes (%) to a 1 percent increase in the policy rate.


4.4.2 Exchange Rate Volatility (Daily Frequency)

Figure 6 depicts the implied IRFs of exchange rate volatility. PanelsA and C show the effects of an unannounced 1 billion USD pur-chase and sale, respectively. In both cases, the CBRT was able tostem volatility, although more so for purchases (5.0 percent) thanfor sales (0.05 percent). But, even though the effect of unannouncedsales is small and short-lived, it is at least contrary to the effectfound if using observed levels of intervention, as shown in panel D,where volatility increases by almost 15 percent.

Foreign exchange intervention through announced purchases of 1billion USD also reduces exchange rate volatility by up to 7 percent,as shown in panel E, but effects subside after fifteen days. Finally,panel G shows that changes in the policy rate have no significantimpact on exchange rate volatility.

4.4.3 Changes in Inflation (Monthly Frequency)

Figure 7 depicts the implied IRFs of changes in inflation. Panel Hshows that an increase of 1 percent in the observed policy rate change(ΔIRIt) has a strong and positive effect on inflation (of 1.0 percent)that lasts for more than fifteen months before the effect subsides.Taken at face value, this result is straightforward evidence of the“price puzzle” in which prices and interest rates are positively cor-related. However, panel G shows that, just like in Romer and Romer(2004), this bias is completely eliminated: an increase of 1 percent inthe policy rate shock lowers inflation by almost 1 percent, and effectsare significant only after the first nine months (i.e., months 9–12).This result is in line with the related literature that find evidence ofalmost a one-year lag delay of interest rates on inflation.28

The remaining panels of figure 7 show that foreign exchangeshocks have no effects on inflation, which is consistent with the factthat almost all interventions were fully sterilized.

4.4.4 Industrial Production Growth (Monthly Frequency)

Figure 8 depicts the implied IRFs of industrial production growth.As shown, only the policy rate shock has a significant effect on

28See, for example, Batini and Nelson (2001), Romer and Romer (2004), andHavranek and Rusnak (2012).


Figure 6. Implied IRFs of Exchange Rate Volatility (Volt)



Figure 7. Implied IRFs of Inflation Changes (Δπt)



Figure 8. Implied IRFs of Industrial Production (ΔIndt)



industrial output. Namely, panel G shows that a 1 percent increasein the policy rate lowers output growth by up to 2 percent duringthe second trimester (months 4–6) following a monetary contraction.This result is line with Romer and Romer (2004), Kilinc and Tunc(2014), and Villamizar-Villegas (2016).

4.4.5 Robustness

In essence, these results differ from the foreign exchange interven-tion literature such as Herrera and Ozbay (2005), Gersl and Holub(2006), Humala and Rodriguez (2010), Echavarria et al. (2013),Adler and Tovar (2014), and Kilinc and Tunc (2014), in that we(i) control for the covariance of simultaneous policies, (ii) includeonly the period in which the CBRT officially adopted an inflation-targeting regime, and (iii) use a different measure of the policy rate.

In terms of the latter (i.e., different measure of policy), we con-duct a robustness exercise presented in appendix 3 in which we com-pare the effects of using the interbank rate on inflation and output.29

As depicted in figure 10 in appendix 3, while the effects of bothour measure of policy and the interbank rate follow similar paths,they greatly differ in the size of the standard errors. Consequently,results using the interbank rate show a more immediate (significant)response of policy. We thus believe that factors related to liquid-ity demand, which in turn are highly correlated with inflation andoutput, can bias the significance level of the obtained estimates.

5. Conclusions

Following the 2001 crisis, the Turkish economy underwent a struc-tural transformation. The Central Bank of the Republic of Turkeywas vested with independence and endowed with the primary objec-tive of achieving and maintaining price stability. In 2002, the CBRTofficially adopted an inflation-targeting regime. Concurrently, inorder to lower exchange rate volatility, the CBRT conducted foreignexchange interventions in one of two ways: (i) through unannounced

29Similar to Kilinc and Tunc (2014), we use the overnight repo rate obtainedfrom the Istanbul Stock Exchange (Borsa Istanbul) as a measure of the interbankrate.


interventions, often infrequent but large, and (ii) through announcedinterventions which consisted of predetermined dates and amounts.

In this paper we study the effects of simultaneous central bankpolicies in a unified framework. Namely, we analyze the effectsof both interest rate and foreign exchange intervention on severalmacroeconomic variables. To this end, we model the undertakingsof monetary authorities (tailored to the various foreign exchangemechanisms of the CBRT) and purge the intended monetary deci-sions from systematic responses of policy. Our investigation confirmssome of the previous findings from the literature but also yields somenew results. For instance, we find that the price puzzle disappearsonce monetary shocks are purged from inflation expectations andglobal financial conditions. Additionally, we find that unannouncedpurchases of foreign currency had a significant effect in reducingexchange rate volatility but appeared to have no effect on exchangerate changes. Announced interventions, on the other hand, did havea significant impact on exchange rate changes and volatility.

Appendix 1. Data Description

Source: Central Bank of the Republic of Turkey during January2002–July 2010.

• Policy instruments of the Central Bank of the Republic ofTurkey (D1t and D2t):– Foreign Exchange Interventions (FXIt), in millions of U.S.

dollars:* Unannounced Purchases and Sales: Exercised through

direct sales and purchases of USD.* Announced Purchases and Sales: Sales and purchases

of USD with predetermined dates and amounts.* Optional Purchases: During auctions of announced pur-

chases and sales, the CBRT may have exceeded thepredetermined amount of FX purchases. There wasa limit on how much authorities could exceed duringthese transactions. The CBRT never optionally exer-cised sales of USD.

– Interest Rate Interventions (IRIt): The policy rate corre-sponded to the central bank’s overnight borrowing rate


between February 20, 2002 and May 16, 2008 (due to theabundant liquidity in the Turkish market); to the overnightlending rate between May 17, 2008 and May 20, 2010 (dueto the liquidity shortage); and to the one-week repo lend-ing rate after May 21, 2010. Frequency corresponds tothe meeting dates of the board of directors. Units are inchanges (%).

• Variables in Xt:– Weekly Depreciation (Δet−5): Weekly (five business days)

exchange rate changes. Daily frequency. Units are in log-differences.

– Monthly Depreciation (Δet−20): Monthly (twenty businessdays) exchange rate changes. Daily frequency. Units are inlog-differences.

– U.S. Federal Funds Rate (Δi∗t ): Self-explanatory. Dailyfrequency. Units are in changes (%).

– Inflation Surprises (πt − π∗t ): Expected inflation corre-

sponds to the one-year-ahead forecasts conducted by theCBRT. Biweekly. Units are in (%).

– Inflation minus Yearly Target (πt − π∗t ): Self-explanatory.

Monthly frequency. Units are in (%).– The S&P Goldman Sachs Commodity Price Index,

Bloomberg Ticker SPGCCI (ΔSPGCCIt): Closing price.Daily frequency. Units are in changes (%).

– VIX Index (V IXt): Closing price. Daily frequency.• Outcome Variables in Yt:

– Nominal Exchange Rate (et): Nominal exchange rate inunits of Turkish lira per unit of U.S. dollar (TRY/USD).Daily frequency. Units are in log-differences.

– Exchange Rate Volatility (Volt): Squared daily exchangerate returns (Δet). Daily frequency.

– Industrial Production Growth (ΔIndt): Industrial pro-duction variation. Monthly frequency. Units are in log-differences.

– Inflation (πt): Yearly changes for the Turkish Consumer’sPrice Index (CPI). Monthly frequency. Units are in changes(%).


Appendix 2. Global Financial Conditions and FXI

Figure 9. VIX and Foreign Exchange Interventions

Appendix 3. Using Alternative Measures of the PolicyRate

Figure 10. Robustness Exercise: Effects of UsingDifferent Measures of Policy

Notes: IRFs A–B denote a response in inflation (panel A) and industrial outputgrowth (panel B) to a 1 percent increase in the policy rate.


Appendix 4. Statistical Properties

Table 10. Elliott-Rothenberg-Stock Test for Unit Root

1% 10%Critical Critical

Variable (up to 28 Lags) t-statistic Value Value

Unannounced FX Purchasest −33.064 −3.480 −2.570Unannounced FX Salest −10.255 −3.480 −2.570Announced FX Purchasest −3.771 −3.480 −2.570Announced FX Salest −10.911 −3.480 −2.570Optional FX Purchasest −3.808 −3.480 −2.570Policy Rate (ΔIRIt) −6.631 −3.480 −2.570Inflation (Δπt) −4.183 −3.480 −2.570Inflation (πt) – Expected (πe

t ) −2.648 −3.480 −2.570Inflation (πt) – Target (π∗

t ) −2.640 −3.480 −2.570Exchange Rate (Δet) −8.201 −3.480 −2.570Monthly Exchange Rate (Δet−20) −6.361 −3.480 −2.570Exchange Rate Volatility (Vol t) −5.916 −3.480 −2.570Industrial Production (ΔIndt) −3.066 −3.480 −2.570U.S. Federal Funds Rate (Δi∗t ) −9.476 −3.480 −2.570Commodity Price Index (ΔSPGCCIt) −7.837 −3.480 −2.570VIX Indext −2.816 −3.480 −2.570

Notes: The minimum lag is determined using the modified Akaike information cri-terion (MAIC). All variables reject the null hypothesis of a unit root at the 1 percentlevel (except for industrial production growth, inflation minus its yearly target, infla-tion minus its yearly forecast, and the VIX index, which reject the null at the 0 percentlevel).

References

Adler, G., and C. E. Tovar. 2011. “Foreign Exchange Intervention: AShield Against Appreciation Winds?” IMF Working Paper No.11/165.

———. 2014. “Foreign Exchange Interventions and Their Impact onExchange Rate Levels.” Monetaria II (1): 1–48.

Akinci, O., O. Y. Culha, U. Ozlale, and G. Sahinbeyoglu. 2006. “TheEffectiveness of Foreign Exchange Interventions under a Floating


Exchange Rate Regime for the Turkish Economy: A Post-CrisisPeriod Analysis.” Applied Economics 38 (12): 1371–88.

Angrist, J. D., and J.-S. Pischke. 2009. Mostly Harmless Economet-rics: An Empiricist’s Companion. Princeton University Press.

Baillie, R. T., and W. P. Osterberg. 1997. “Why Do Central BanksIntervene?” Journal of International Money and Finance 16 (6):909–19.

Barr, D. R., and E. T. Sherrill. 1999. “Mean and Variance of Trun-cated Normal Distributions.” American Statistician 53 (4): 357–61.

Batini, N., and E. Nelson. 2001. “The Lag from Monetary PolicyActions to Inflation: Friedman Revisited.” International Finance4 (3): 381–400.

Blanchard, O. 2013. “Monetary Policy Will Never Be the Same.”Technical Report. IMF Blog.

Bonser-Neal, C., and G. Tanner. 1996. “Central Bank Interventionand the Volatility of Foreign Exchange Rates: Evidence from theOptions Market.” Journal of International Money and Finance15 (6): 853–78.

Canzoneri, M., and R. Cumby. 2013. “Optimal Foreign ExchangeIntervention in an Inflation Targeting Regime: Some CautionaryTales.” Open Economies Review 25 (3): 429–50.

Chang, Y., and S. J. Taylor. 1998. “Intraday Effects of ForeignExchange Intervention by the Bank of Japan.” Journal of Inter-national Money and Finance 17 (1): 191–210.

Cohen, A. C. 1949. “On Estimating the Mean and Standard Devia-tion of Truncated Normal Distributions.” Journal of the Ameri-can Statistical Association 44 (248): 518–25.

Disyatat, P., and G. Galati. 2007. “The Effectiveness of ForeignExchange Intervention in Emerging Market Countries: Evidencefrom the Czech Koruna.” Journal of International Money andFinance 26 (3): 383–402.

Domac, I., and A. Mendoza. 2004. “Is There Room for ForeignExchange Interventions under an Inflation Targeting Frame-work? Evidence from Mexico and Turkey.” Policy ResearchWorking Paper No. 3288, World Bank.


Dominguez, K. M. 1993. “Does Central Bank Intervention Increasethe Volatility of Foreign Exchange Rates?” NBER WorkingPaper No. 4532.

Dominguez, K. M., and J. Frankel. 1993. “Does Foreign-ExchangeIntervention Matter? The Portfolio Effect.” American EconomicReview 83 (5): 1356–69.

Echavarria, J. J., L. F. Melo, S. Tellez, and M. Villamizar. 2013.“The Impact of Pre-announced Day-to-Day Interventions on theColombian Exchange Rate.” BIS Working Paper No. 428.

Eichengreen, B. 1994. International Monetary Arrangements for the21st Century. Washington, DC: Brookings Institution.

Fatum, R., and M. Hutchison. 1999. “Is Intervention a Signalof Future Monetary Policy? Evidence from the Federal FundsFutures Market.” Journal of Money, Credit and Banking 31 (1):54–69.

———. 2003. “Is Sterilised Foreign Exchange Intervention EffectiveAfter All? An Event Study Approach.” Economic Journal 113(487): 390–411.

Fischer, S. 2001a. “Exchange Rate Regimes: Is the Bipolar ViewCorrect?” Journal of Economic Perspectives 15 (2): 3–24.

———. 2001b. “The International Financial System: Crises andReform.” The Robbins Lectures at the London School of Eco-nomics, October 29–31.

Gersl, A., and T. Holub. 2006. “Foreign Exchange InterventionsUnder Inflation Targeting: The Czech Experience.” Contempo-rary Economic Policy 24 (4): 475–91.

Guimaraes, R., and C. Karacadag. 2004. “The Empirics of ForeignExchange Intervention in Emerging Market Countries: The Casesof Mexico and Turkey.” IMF Working Paper No. 04/123.

Havranek, T., and M. Rusnak. 2012. “Transmission Lags of Mon-etary Policy: A Meta-Analysis.” Working Paper No. 2012/10,Research Department, Czech National Bank.

Herrera, A. M., and P. Ozbay. 2005. “A Dynamic Model of CentralBank Intervention.” Working Paper No. 0501, Central Bank ofthe Republic of Turkey.

Humala, A., and G. Rodriguez. 2010. “Foreign Exchange Interven-tion and Exchange Rate Volatility in Peru.” Applied EconomicsLetters 17 (15): 1485–91.


Humpage, O. F. 1999. “U.S. Intervention: Assessing the Probabil-ity of Success.” Journal of Money, Credit and Banking 31 (4):731–47.

Jorda, O. 2005. “Estimation and Inference of Impulse Responses byLocal Projections.” American Economic Review 95 (1): 161–82.

Kamil, H. 2008. “Is Central Bank Intervention Effective Under Infla-tion Targeting Regimes? The Case of Colombia.” IMF WorkingPaper No. 08/88.

Kara, H. 2013. “Safeguarding Macroeconomic and Price Stabilityamid Capital Flow Volatility.” Paper prepared for the BIS Meet-ing of Central Bank Chief Economists, Basel, Switzerland, March2013.

Kearns, J., and R. Rigobon. 2002. “Identifying the Efficacy of Cen-tral Bank Interventions: The Australian Case.” NBER WorkingPaper No. 9062.

Kilinc, M., and C. Tunc. 2014. “Identification of Monetary PolicyShocks in Turkey: A Structural VAR Approach.” Working PaperNo. 1423, Research and Monetary Policy Department, CentralBank of the Republic of Turkey.

Kim, S.-J., T. Kortian, and J. Sheen. 2000. “Central Bank Interven-tion and Exchange Rate Volatility: Australian Evidence.” Jour-nal of International Financial Markets, Institutions and Money10 (3–4): 381–405.

Menkhoff, L. 2013. “Foreign Exchange Intervention in EmergingMarkets: A Survey of Empirical Studies.” World Economy 36(9): 1187–1208.

Neil, B., and J.-F. Fillion. 1999. “An Intraday Analysis of the Effec-tiveness of Foreign Exchange Intervention.” Working Paper No.99-4, Bank of Canada.

Obstfeld, M., and K. Rogoff. 1995. “The Mirage of Fixed ExchangeRates.” Journal of Economic Perspectives 9 (4): 73–96.

Rey, H. 2015. “Dilemma not Trilemma: The Global Financial Cycleand Monetary Policy Independence.” NBER Working Paper No.21162.

Rincon, H., and J. Toro. 2010. “Are Capital Controls and CentralBank Intervention Effective?” Borradores de Economia No. 625,Banco de la Republica de Colombia.


Romer, C. D., and D. H. Romer. 2004. “A New Measure of Mone-tary Shocks: Derivation and Implications.” American EconomicReview 94 (4): 1055–84.

Rosenbaum, S. 1961. “Moments of a Truncated Bivariate NormalDistribution.” Journal of the Royal Statistical Society: Series B(Methodological) 23 (2): 405–408.

Taylor, M. P. 2004. “Is Official Exchange Rate Intervention Effec-tive?” Economica 71 (281): 1–11.

Villamizar-Villegas, M. 2016. “Identifying the Effects of Simultane-ous Monetary Policy Shocks.” Contemporary Economic Policy34 (2): 268–96.

Villamizar-Villegas, M., and D. Perez-Reyna. 2017. “A TheoreticalApproach to Sterilized Foreign Exchange Intervention.” Journalof Economic Surveys 31 (1): 343–65.

The Effects of Monetary and Exchange Rate Policy Shocks ... · PDF fileUribe, Juan Pablo...

Documents

Transcript of The Effects of Monetary and Exchange Rate Policy Shocks ... · PDF fileUribe, Juan Pablo...